srm: fix bring-online rescheduling on restart

Several states of BringOnlineFileRequest were not handled correctly. This patch fixes this. Target: master Patch: http://rb.dcache.org/r/6807/ Acked-by: Gerd Behrmann
dCache · Mar 31, 2014 · 780ae6e · 780ae6e
1 parent 268de60
commit 780ae6e
Show file tree

Hide file tree

Showing 4 changed files with 31 additions and 7 deletions.
diff --git a/modules/srm-server/src/main/java/org/dcache/srm/request/BringOnlineFileRequest.java b/modules/srm-server/src/main/java/org/dcache/srm/request/BringOnlineFileRequest.java
@@ -383,9 +383,28 @@ public final void run() throws NonFatalJobFailure, FatalJobFailure {
     protected void onSrmRestartForActiveJob(Scheduler scheduler)
             throws IllegalStateTransition
     {
-        // Simply reschedule it (FIXME: this doesn't always work due to
-        // illegal state transitions)
-        scheduler.schedule(this);
+        State state = getState();
+
+        switch (state) {
+        case ASYNCWAIT:
+        case RETRYWAIT:
+            // FIXME: we should log the SRM restart in the job's history.
+            scheduler.schedule(this);
+            break;
+
+        case PRIORITYTQUEUED:
+        case TQUEUED:
+        case RUNNING:
+            setState(State.RESTORED, "Rescheduled after SRM service restart");
+            scheduler.schedule(this);
+            break;
+
+        // All other states are invalid.
+        default:
+            setState(State.FAILED, "Invalid state (" + state + ") detected " +
+                    "after SRM service restart");
+            break;
+        }
     }
 
     public void askFileId() throws NonFatalJobFailure, FatalJobFailure {

diff --git a/modules/srm-server/src/main/java/org/dcache/srm/request/Job.java b/modules/srm-server/src/main/java/org/dcache/srm/request/Job.java
@@ -360,11 +360,13 @@ private boolean isValidTransition(State currentState, State newState)
         case TQUEUED:
             return newState == State.CANCELED
                     || newState == State.FAILED
-                    || newState == State.RUNNING;
+                    || newState == State.RUNNING
+                    || newState == State.RESTORED;
         case PRIORITYTQUEUED:
             return newState == State.CANCELED
                     || newState == State.FAILED
-                    || newState == State.RUNNING;
+                    || newState == State.RUNNING
+                    || newState == State.RESTORED;
         case RUNNING:
             return newState == State.CANCELED
                     || newState == State.FAILED
@@ -373,7 +375,8 @@ private boolean isValidTransition(State currentState, State newState)
                     || newState == State.RQUEUED
                     || newState == State.READY
                     || newState == State.DONE
-                    || newState == State.RUNNINGWITHOUTTHREAD;
+                    || newState == State.RUNNINGWITHOUTTHREAD
+                    || newState == State.RESTORED;
         case ASYNCWAIT:
             return newState == State.CANCELED
                     || newState == State.FAILED
@@ -399,6 +402,8 @@ private boolean isValidTransition(State currentState, State newState)
             return newState == State.CANCELED
                     || newState == State.FAILED
                     || newState == State.DONE;
+        case RESTORED:
+            return newState == State.PRIORITYTQUEUED;
         case FAILED:
         case DONE:
         case CANCELED:

diff --git a/modules/srm-server/src/main/java/org/dcache/srm/scheduler/Scheduler.java b/modules/srm-server/src/main/java/org/dcache/srm/scheduler/Scheduler.java
@@ -251,6 +251,7 @@ public void schedule(Job job)
                     job.setState(State.FAILED, "Site busy: too many queued requests.");
                 }
                 break;
+            case RESTORED:
             case ASYNCWAIT:
             case RETRYWAIT:
             case RUNNINGWITHOUTTHREAD:

diff --git a/modules/srm-server/src/main/java/org/dcache/srm/scheduler/State.java b/modules/srm-server/src/main/java/org/dcache/srm/scheduler/State.java
@@ -112,7 +112,6 @@ public enum State {
 
     /** A resource limitation or some failure prevented job from completing successfully. */
     FAILED         ("Failed"         ,11, true),
-    @Deprecated // Kept for compatibility with old databases
     RESTORED       ("Restored"       ,12),
 
     /** Job has triggered a third-party copy. */